Numerical Simulation for Fredholm Integral Equation of the Second Kind

Peng Guo^{1,2}

Show more

1. Introduction

Integral equation is a very important tool, it has been more and more widely used in Mechanics, Mathematics, Thermodynamics, Physics and other fields in recent decades. As early as 1823, Norwegian mathematician N.H. Abel has obtained the integral equation of the following form when he studied the tautochrone problem. Now, the integral equation is also known as the Abel equation,

${\int}_{a}^{x}}\frac{G\left(x\mathrm{,}t\right)u\left(t\right)\text{d}t}{{\left(x-t\right)}^{1-\alpha}}=f\left(x\right)\mathrm{.$ (1)

This integral equation can be regarded as a special case of Volterral’s integral equation, which can also be transformed into a fractional differential equation. The theory of integral equation is established mainly by Fredholm and Volterra in the late 19^{th} century. Their work profoundly influenced the study of integral equations in the 20^{th} century. Many mathematical models in science and engineering can be described by integral equation model, such as anomalous diffusion problem, population prediction model, biological population ecological model, nerve pulse propagation, medical scanning, viscoelastic material simulation, heat conduction problem with memory material, *etc.*, one can refer to [1] - [6].

In this work, we will focus on the Fredholm integral equation of the second kind,

$y\left(x\right)=\lambda {\displaystyle {\int}_{a}^{b}}\text{\hspace{0.05em}}\text{\hspace{0.05em}}\varphi \left(x\mathrm{,}t\right)y\left(t\right)\text{d}t+f\left(x\right)\mathrm{,}$ (2)

where $\varphi \left(x\mathrm{,}t\right)$ is the kernal of the integral equation, $\lambda $ is the coefficient of the integral item, $f\left(x\right)$ is a free item. $y\left(x\right)$ is the unknown function, a and b are the nodes of the integral interval.

As is well known, the analytical solution of the integral equation is hard to get. Only in some special cases, such as the integral part contains degenerate kernal, the analytical solution can be presented. The numerical method is a very useful tool for solving the integral equation which is difficult to obtain the analytic solution, even the integral equation whose analytic solution does not exist at all. From different research perspectives, many scholars have proposed various numerical methods, such as finite difference method, Galerkin method, finite element method, Nyström method, allocation method, successive approximation method, *etc.*

Guo *et al.* used a finite difference scheme for solving the nonlinear time-fractional partial integro-differential equation [7]. Assari *et al.* presented a discrete Galerkin method for solving Fredholm integral equations of the second kind with logarithmic kernels [8]. Brambilla *et al.* described the implementation of a one-dimensional integral equation in a finite-element model [9]. Guoqiang and Jiong got a Nystrom solution for two-dimensional nonlinear Fredholm integral equations of the second kind, and gave an error analysis for this method [10]. Chen and Tang analyzed the convergence of the Jacobi spectral-collocation methods for Volterra integral equations [11]. Maleknejad and Sohrabi provided a numerical solution of Fredholm integral equations of the first kind by using Legendre wavelets [12]. Yusufoglu and Erbas obtained a numerical solution for Fredholm-Volterra type integral equations by interpolation and quadrature rules [13]. And other one can refer to [14] - [21].

2. Preliminaries

In this part, the properties of the Legendre polynomials will be shown. And the zeros of the Legendre polynomials will be used to construct a numerical algoruthm in the following part.

Definition 1. [22] In the interval $\left[-\mathrm{1,1}\right]$, with the weight $\rho \left(x\right)=1$, the polynomials with the following relations are called Legendre polynomicals.

${P}_{0}\left(x\right)=1,{P}_{n}\left(x\right)=\frac{1}{{2}^{n}n!}\frac{{\text{d}}^{n}\left[{\left({x}^{2}-1\right)}^{n}\right]}{\text{d}{x}^{n}},$ (3)

where $n=1,2,3,\cdots $.

From the above relations, it is clear that the coefficient of Legendre Polynomials ${P}_{n}\left(x\right)$ is $\frac{\left(2n\right)\mathrm{!}}{{2}^{n}{\left(n\mathrm{!}\right)}^{2}}$.

Property 1.

${\int}_{-1}^{1}}\text{\hspace{0.05em}}\text{\hspace{0.05em}}{P}_{m}\left(x\right){P}_{n}\left(x\right)\text{d}x=\{\begin{array}{ll}\frac{2}{2n+1},\hfill & m=n,\hfill \\ 0,\hfill & m\ne n.\hfill \end{array$ (4)

Property 2.

${P}_{n}\left(-x\right)={\left(-1\right)}^{n}{P}_{n}\left(x\right)\mathrm{.}$ (5)

Property 3.

${P}_{n+1}\left(x\right)=\frac{2n+1}{n+1}x{P}_{n}\left(x\right)-\frac{n}{n+1}{P}_{n-1}\left(x\right),$ (6)

where $n=1,2,3,\cdots $.

Similarly, the zeros of Chebyshev polynomials and other special polynomials also can be used to cope with this problem, one can refer to [22]. Legendre polynomials are chosen here mainly because the weight coefficient is 1, which provides a great convenience for calculation.

3. Nunmerical Algorithm and Error Analysis for Fredholm Integral Equation of the Second Kind

Let’s consider the Gauss-Lobatto formula in the interval $\left[-\mathrm{1,1}\right]$

${\int}_{-1}^{1}}\text{\hspace{0.05em}}\text{\hspace{0.05em}}g\left(t\right)\text{d}t={\displaystyle \underset{j=1}{\overset{n}{\sum}}}\text{\hspace{0.05em}}\text{\hspace{0.05em}}{\omega}_{j}g\left({t}_{j}\right)+{R}_{n}\left[g\right],$ (7)

where the coefficients ${\omega}_{j},j=1,2,3,\cdots $ satisfy the following relations

${\omega}_{j}=\{\begin{array}{ll}\frac{2}{n\left(n+1\right)},\hfill & {t}_{j}=\pm 1,\hfill \\ \frac{2}{n\left(n+1\right){P}_{n}^{2}\left({t}_{j}\right)},\hfill & \text{else},\hfill \end{array}$ (8)

where ${P}_{n}\left(x\right)$ are the Legendre polynomials and ${t}_{j},j=1,2,3$ are the zeros of ${{P}^{\prime}}_{n}\left(x\right)$. The trunction error is

${R}_{n}\left[g\right]=-\frac{\left(n+1\right){n}^{3}{2}^{2n+1}{\left(\left(n-1\right)!\right)}^{4}{g}^{\left(2n\right)}\left(\xi \right)}{\left(2n+1\right){\left(\left(2n\right)!\right)}^{3}},\xi \in \left(-1,1\right).$ (9)

If the interval is $\left[a\mathrm{,}b\right]$, one should make the following substitution $t=\frac{\left(b-a\right)s+b+a}{2}$. Then the interval $\left[a\mathrm{,}b\right]$ can be mapped into $\left[-\mathrm{1,1}\right]$. And the Fredholm integral equation can be equivalently transformed to the following form

$y\left(x\right)=\frac{\lambda \left(b-a\right)}{2}{\displaystyle {\int}_{-1}^{1}}\text{\hspace{0.05em}}\text{\hspace{0.05em}}\varphi \left(x,\frac{\left(b-a\right)s+b+a}{2}\right)y\left(\frac{\left(b-a\right)s+b+a}{2}\right)\text{d}s+f\left(x\right),$ (10)

where $x\in \left[a\mathrm{,}b\right]$, $\frac{\left(b-a\right)s+b+a}{2}\in \left[a\mathrm{,}b\right]$.

For the Fredholm integral equation of the second kind, the hardest part to deal with is the integral term. The quadrature formula was applied to cope with the integral term. i.e.

$y\left(x\right)=\stackrel{\u02dc}{\lambda}{\displaystyle \underset{j=1}{\overset{n}{\sum}}}\text{\hspace{0.05em}}\text{\hspace{0.05em}}{\omega}_{j}\varphi \left(x,{x}_{j}\right)y\left({x}_{j}\right)+f\left(x\right)+{\stackrel{\u02dc}{R}}_{n},$ (11)

where ${x}_{j}=\frac{\left(b-a\right){s}_{j}+b+a}{2},j=1,2,3,\cdots $ are the nodes, ${\omega}_{j},j=1,2,3,\cdots $ are the coefficients, ${\stackrel{\u02dc}{R}}_{n}$ is the truncation error for the quadrature formula,

${\stackrel{\u02dc}{R}}_{n}=-\frac{\lambda \left(n+1\right){n}^{3}{\left(b-a\right)}^{2n+1}{\left(\left(n-1\right)!\right)}^{4}{G}^{\left(2n\right)}\left(\xi \right)}{\left(2n+1\right){\left(\left(2n\right)!\right)}^{3}},\xi \in \left(-1,1\right),$ (12)

where $G\left(s\right)=\frac{\lambda \left(b-a\right)}{2}\varphi \left(x\mathrm{,}\frac{\left(b-a\right)s+b+a}{2}\right)y\left(\frac{\left(b-a\right)s+b+a}{2}\right)$.

If we denote ${y}_{i}=y\left({x}_{i}\right)$, ${f}_{i}=f\left({x}_{i}\right)$, ${\varphi}_{ij}=\varphi \left({x}_{i},{x}_{j}\right)$, then the following algorithm will be proposed.

${y}_{i}=\stackrel{\u02dc}{\lambda}{\displaystyle \underset{j=1}{\overset{n}{\sum}}}\text{\hspace{0.05em}}\text{\hspace{0.05em}}{\omega}_{j}{\varphi}_{ij}{y}_{j}+{f}_{i}.$ (13)

The coefficient matrix of the above system can be written in detailed as

$C=\left(\begin{array}{cccc}\stackrel{\u02dc}{\lambda}{\omega}_{1}{\varphi}_{11}& \stackrel{\u02dc}{\lambda}{\omega}_{2}{\varphi}_{12}& \cdots & \stackrel{\u02dc}{\lambda}{\omega}_{n}{\varphi}_{1n}\\ \stackrel{\u02dc}{\lambda}{\omega}_{1}{\varphi}_{21}& \stackrel{\u02dc}{\lambda}{\omega}_{2}{\varphi}_{22}& \cdots & \stackrel{\u02dc}{\lambda}{\omega}_{n}{\varphi}_{2n}\\ \vdots & \vdots & \ddots & \vdots \\ \stackrel{\u02dc}{\lambda}{\omega}_{1}{\varphi}_{n1}& \stackrel{\u02dc}{\lambda}{\omega}_{2}{\varphi}_{n2}& \cdots & \stackrel{\u02dc}{\lambda}{\omega}_{n}{\varphi}_{nn}\end{array}\right).$ (14)

Therefore, if $det\left(I-C\right)\ne 0$, one can get the unique approximate solution.

4. Nunmerical Examples

In this part, the proposed algorithm will be employed to solve two Fredholm integral equations of the second kind. To show the efficiency of the algorithm we proposed the numerical results will be compared with the exact solution.

Example 1. Consider the following Fredholm integral equation of the second kind

$y\left(x\right)=-{\displaystyle {\int}_{0}^{1}}\text{\hspace{0.05em}}\text{\hspace{0.05em}}x{\text{e}}^{t}y\left(t\right)\text{d}t+{\text{e}}^{-x},$ (15)

where $x\in \left[\mathrm{0,1}\right]$, and the exact solution is $y\left(x\right)={\text{e}}^{-x}-\frac{x}{2}$.

Figure 1 shows the numerical and analytical solution of Example 1. One can find that the numerical solution and the exact solution are in good agreement. Figure 2 shows the errors at different nodes in the interval
$\left[\mathrm{0,1}\right]$. The error bound is about 10^{−}^{9}, when
$n=4$.

When we choose different collocation nodes, i.e. $n=2,3,4,5,6$. Table 1 shows that the errors at different points in the interval $\left[\mathrm{0,1}\right]$ for Example 1. One can find that the more nodes we use, the better accuracy we get.

Example 2. Consider the following Fredholm integral equation of the second kind

$y\left(x\right)=\frac{2}{\text{e}-1}{\displaystyle {\int}_{0}^{1}}\text{\hspace{0.05em}}\text{\hspace{0.05em}}{\text{e}}^{x}y\left(t\right)\text{d}t-{\text{e}}^{x},$ (16)

where $x\in \left[\mathrm{0,1}\right]$, and the exact solution is $y\left(x\right)={\text{e}}^{x}$.

Figure 1. The exact solution and numerical solution (n = 4).

Figure 2. The error of the exact solution and the numerical solution (n = 4).

Comparison of the numerical and analytical solutions of Example 2 is illustrated in Figure 3. Figure 4 reveals the absolute error of the numerical and analytical solutions for Example 2. Figure 1 and Figure 2 show a good performance of the algorithm we proposed for solving the Fredholm intergral euation of the second kind.

From Table 2, it can be seen that the numerical results with the method we proposed are in excellent agreement with the exact solution. When the collocation nodes
$n=6$, the absolute error is 10^{−}^{16}, which means a high precision of the method we used.

Table 1. The errors of Example 1.

Figure 3. The exact solution and numerical solution (n = 5).

Figure 4. The error of the exact solution and the numerical solution (n = 5).

Table 2. The errors of Example 2.

5. Conclusion

In this work, we study the Fredholm integral equation of the second kind. For the integral term of the equation, we use Gauss-Lobatto quadrature formula to cope with it. We give the numerical algorithm and truncation error for the algorithm we used. Furthermore, we give the existence condition of the algorithm for Fredholm integral equation we discussed. Finally, two numerical examples are shown to verify the efficiency and accuracy of the algorithm.

Acknowledgements

This research was funded by the Humanity and Social Science Youth foundation of Ministry of Education (No. 18YJC630120), the Applied Mathematics of Shanghai DianJi University (No. 16JCXK02) and the Key Projects of Shanghai Soft Science Research Program (No. 18692106600).

References

[1] Magin, R.L., Abdullah, O., Baleanu, D. and Zhou, X.J. (2008) Anomalous Diffusion Expressed through Fractional Order Differential Operators in the Bloch-Torrey Equation. Journal of Magnetic Resonance, 190, 255-270.

https://doi.org/10.1016/j.jmr.2007.11.007

[2] Yla-Oijala, P. and Taskinen, M. (2005) Application of Combined Field Integral Equation for Electromagnetic Scattering by Dielectric and Composite Objects. IEEE Transactions on Antennas and Propagation, 53, 1168-1173.

https://doi.org/10.1109/TAP.2004.842640

[3] Monje-Real, A. and de la Rubia, V. (2020) Electric Field Integral Equation Fast Frequency Sweep for Scattering of Nonpenetrable Objects via the Reduced-Basis Method. IEEE Transactions on Antennas and Propagation, 68, 6232-6244.

https://doi.org/10.1109/TAP.2020.2992882

[4] Tanaka, M. and Tanaka, K. (2019) Method of Moments Based on Electric Field Integral Equation for Three-Dimensional Metallic Waveguide: Single Mode Waveguide. IEICE Transactions on Electronics, E102, 30-37.

https://doi.org/10.1587/transele.E102.C.30

[5] Lindgren, E.B., Stace, A.J., Etienne, P., et al. (2018) An Integral Equation Approach to Calculate Electrostatic Interactions in Many-Body Dielectric Systems. Journal of Computational Physics, 371, 712-731.

https://doi.org/10.1016/j.jcp.2018.06.015

[6] Meral, F.C., Royston, T.J. and Magin, R. (2010) Fractional Calculus in Viscoelasticity: An Experimental Study. Communications in Nonlinear Ence & Numerical Simulation, 15, 939-945.

https://doi.org/10.1016/j.cnsns.2009.05.004

[7] Guo, J., Xu, D. and Qiu, W. (2019) A Finite Difference Scheme for the Nonlinear Time-Fractional Partial Integro-Differential Equation. Mathematical Methods in the Applied Ences, 43, 1-21.

https://doi.org/10.1002/mma.6128

[8] Assari, P., Adibi, H. and Dehghan, M. (2014) A Meshless Discrete Galerkin (mdg) Method for the Numerical Solution of Integral Equations with Logarithmic Kernels. Journal of Computational & Applied Mathematics, 267, 160-181.

https://doi.org/10.1016/j.cam.2014.01.037

[9] Brambilla, R., Grilli, F., Martini, L., Sirois, F. (2008) Integral Equations for the Current Density in Thin Conductors and Their Solution by the Finite-Element Method. Superconductor Ence and Technology, 21, Article ID: 105008.

https://doi.org/10.1088/0953-2048/21/10/105008

[10] Han, G.Q. and Wang, J. (2001) Extrapolation of Nystrom Solution for Two Dimensional Nonlinear Fredholm Integral Equations. Journal of Computational and Applied Mathematics, 134, 259-268.

https://doi.org/10.1016/S0377-0427(00)00553-7

[11] Chen, Y. and Tang, T. (2009) Convergence Analysis of the Jacobi Spectral-Collocation Methods for Volterra Integral Equations with a Weakly Singular Kernel. Mathematics of Computation, 79, 147-167.

https://doi.org/10.1090/S0025-5718-09-02269-8

[12] Maleknejad, K. and Sohrabi, S. (2007) Numerical Solution of Fredholm Integral Equations of the First Kind by Using Legendre Wavelets. Applied Mathematics and Computation, 186, 836-843.

https://doi.org/10.1016/j.amc.2006.08.023

[13] Yusufoglu, E. and Erbas, B. (2008) Numerical Expansion Methods for Solving Fredholm-Volterra Type Linear Integral Equations by Interpolation and Quadrature Rules. Kybernetes, 37, 768-785.

https://doi.org/10.1108/03684920810876972

[14] Wazwaz, A.M. (2011) The Regularization Method for Fredholm Integral Equations of the First Kind. Computers & Mathematics with Applications, 61, 2981-2986.

https://doi.org/10.1016/j.camwa.2011.03.083

[15] Wang, G. (2015) Application of Wavelets on the Interval to Numerical Analysis of Integral Equations in Electromagnetic Scattering Problems. International Journal for Numerical Methods in Engineering, 40, 1-13.

https://doi.org/10.1002/(SICI)1097-0207(19970115)40:1<1::AID-NME44>3.0.CO;2-V

[16] Sidorov, N.A. and Sidorov, D.N. (2010) Solving the Hammerstein Integral Equation in the Irregular Case by Successive Approximations. Siberian Mathematical Journal, 51, 325-329.

https://doi.org/10.1007/s11202-010-0033-4

[17] Golovach, G.P. and Kalaida, A.F. (2012) Convergence Rate of the Successive Approximation Method for the Solution of Nonlinear Volterra Type Integral Equations. Neurology, 78, 146.

[18] Young, P., Hao, S. and Martinsson, P.G. (2012) A High-Order Nystrom Discretization Scheme for Boundary Integral Equations Defined on Rotationally Symmetric Surfaces. Journal of Computational Physics, 231, 4142-4159.

https://doi.org/10.1016/j.jcp.2012.02.008

[19] Beams, N.N., Andreas, K. and Olson, L.N. (2018) High-Order Finite Element-Integral Equation Coupling on Embedded Meshes. Journal of Computational Physics, 375, 295-1313.

https://doi.org/10.1016/j.jcp.2018.08.032

[20] Bao, W. (2019) Long Time Error Analysis of Finite Difference Time Domain Methods for the Nonlinear Klein-Gordon Equation with Weak Nonlinearity. Communications in Computational Physics, 26, 1307-1334.

https://doi.org/10.4208/cicp.2019.js60.03

[21] Bin, J.H. (2018) A Gaussian Radial Basis Function-Finite Difference Technique to Simulate the Hcir Equation. Journal of Computational & Applied Mathematics, 347, 181-195.

[22] Wang, Z.X. and Guo, D.R. (1982) Special Functions. World Scientific Press, Singapore.